WeftWight language · C4AI v0.2

Stories with room to diverge, return, and remain coherent.

WeftWight is private, setting-neutral narrative machinery for adventures people design together. It is not a universal fiction ontology and it is not a genre product. It supplies a small language for places, moving cohorts, divergence, optional activities, and accountable AI assistance; a setting supplies the meaning.

Core promise: authored intent, bounded inputs, and exact versioned machinery can reduce to a reproducible result. An AI can help describe, organize, and propose; it does not become the resolver, the authority, or the source of truth.

1. The design stance

Stories are allowed to be fantasy quests, space journeys, romances, mysteries, domestic life, historical imaginings, or something that has no established genre name. WeftWight therefore avoids treating labels such as race, class, weapon, planet, magic, love interest, or job as engine-wide facts. They are setting-owned words and fields.

The engine language concerns only relationships that repeat across settings: a place can contain another place, a declared route can connect places, a cohort can travel, a cohort can split and rejoin under declared rules, an optional activity can create a bounded signal, and a later system can resolve authorized consequences deterministically.

2. Narrative shape: Arc, Thread, and Junction

Narrative Arc

A Narrative Arc is the legible shape of an adventure over time: pressure, choice, consequence, revelation, respite, and continuation. The current Arc viewer draws this as a spatial, organic curve because that makes character positions and authored nodes easy to inspect. The curve is presentation and evidence; it is not itself a travel rule or a universal plot template.

Thread

A Thread is the mechanical form of a splintered arc: an independently advancing cohort carrying a domain, capabilities, a narrative tick, and a travel rate. It answers a concrete question that a drawn fork alone cannot: who went left, who went right, what each group carried, where each ended, and when they arrived.

Junction

A Junction is the declared operation that creates, coordinates, or reunites Threads. It produces versioned resolver evidence rather than relying on an AI’s interpretation of a diagram.

A valid split cannot silently duplicate a character, lose someone at rejoin, invent a capability, or flatten away the slower branch’s time. A branch is allowed to be emotionally or narratively strange; it is not allowed to be mechanically ambiguous.

3. Environmental coherence: Environment, Domain, Route, Projection

Environment and Domain

An Environment is a named spatial graph, not merely a painted map. It has a primary projection and declared contained Domains. A Domain is a bounded place with a setting-owned category and optional parent. That permits a world to contain a region, a region to contain a camp, a camp to contain a dungeon, or a star system to contain a ship, deck, and hull without declaring any of those words universal.

Projection

surface-2d is a grounded X/Z presentation with Y explicit for elevation: land, sea-top, tactical board, town, or ship deck. volume-3d allows full three-dimensional spatial placement: orbital travel, air combat, underwater movement, zero-g interiors, and similar spaces. Either kind of Environment may contain Domains with the other projection. A 2D world can hold a 3D cave; a 3D system can hold a 2D cabin.

Route

A Route is an explicit directed connection between Domains. It has travel units, optional availability in narrative time, and exact setting-owned capability requirements. Proximity on a map does not infer a route. A pretty path, terrain pixel, AI suggestion, or narrative desire does not make movement legal.

This is the foundation for environmental familiarity: an Arc can project onto a map or a volume, contour around authored topology, and enter a nested focused environment without losing the thread’s evidence. Environment-to-Environment transitions are a planned authoring surface; they must bind explicit source, target, route, and receipt rather than being treated as a canvas click.

4. Influence and narrative spheres

A Narrative Sphere is the design term for a bounded, often intentionally hidden, contextual pressure around an area, asset, time, relationship, or node. It can express ideas such as an abandoned road feeling watched, a distant planet offering a research opportunity, a camp becoming safer after preparation, or a relationship becoming more fragile under a secret. Spheres help an author build familiarity and layered possibility without turning every map point into a visible quest marker.

Design direction — not yet a public resolver contract

A sphere must eventually be declared, versioned, scoped, and resolved through reviewed deterministic machinery. It must not be a browser-side score, an AI-originated mechanics change, or a covert way to mint influence. Positive influence, eligibility, locks, visibility, canon, and publication remain separately authorized concerns.

5. Fringe: optional life around the caravan

Fringe is gameplay at the edge of the main travel or arc: camp work, cooking, hunting, alchemy, repairs, navigation research, ship maintenance, or any setting-specific activity that prepares a group for what comes next. A Fringe Module attaches to a Thread, cohort, Domain, asset, or time window. It is a modular attachment point, not uploaded code.

The current closed interaction grammar is deliberately small: click-sequence, timed-choice, trace-path, and inventory-selection. A module can emit bounded signals such as preparation, navigation, resource, condition, time, or flavor. It cannot execute JavaScript, call a service, claim a locally completed mini-game earned durable influence, or conceal mechanics inside art.

Presentation may use an approved art recipe reference so a camp stump, forge, hull panel, or orchard has a coherent visual style. Art and description make an activity inviting; the reviewed declarative module and host rules decide what is mechanically possible.

6. C4AI: Context For AI

C4AI is the general practice of copying selected, bounded context so an AI teammate can understand a task and begin useful work quickly. It is not owned by WeftWight, an autonomous game master, or a provider integration. Within WeftWight, C4AI handoffs can point to this public language repository for exact terms while retaining a hard separation between suggestions and authority.

The copy/paste route is permanent by design. It lets a person use a preferred subscription, local model, or private relationship without giving WeftWight an API key or the whole project. A later direct connection must keep provider credentials in a private local/host connector, show data and billing scope, carry budgets and audit records, and never put a key in browser storage.

7. The C4AI context stack

Each request is organized rather than composed as one undifferentiated prompt:

1. System and schema rules — exact response contract, forbidden actions, and authority boundary.
2. Active scope — the selected stage or field references, current values, and author notes.
3. World context — concise upstream decisions: premise/tags, Environment/start, characters, and primary gear as relevant.
4. Side context — advisory available tags and project vocabulary, never other authors’ private data.
5. Documentation — this versioned public registry, document hashes, and small inline constraints for offline use.

The field-selection direction is important: a future floating C4AI composer lets an author click the exact form fields to load, attach a note to each selection, and receive a patch proposal. A response that omits a field leaves that field unchanged. The current private v0.1 UI uses stage-level selection; the selective field composer and local Loom subscriber integration are planned, private UI work—not a public web endpoint.

When an app supplies an exact response contract, the AI must return only that shape and bind it to the current request. Outside an app, use the context starter and ask for discussion or a reviewable plan. An AI may interpret, express, and propose; it may not change signal strength, eligibility, locks, provenance, visibility, projected state, canon, publication, or provider authority.

8. Determinism, evidence, and trust

WeftWight separates what a person sees from what the system can prove. Contracts and pure kernels consume complete immutable inputs and return deterministic results, reason codes, and content hashes. Decisions and consumption links are append-only so a later derivation can replay exact versioned inputs. Browser views receive a minimized projection; they are never the intellectual-property or authority boundary.

Authentication is not attribution. A display name, project path, local access code, Git identity, or AI chat subscription does not prove ownership or grant a role. A public site is a documentation surface only. It must never carry the private resolver, provider routing, raw traces, credentials, private setting bundles, or operator diagnostics.

9. What exists today and what follows

Fantasy and space are calibration settings, not destinations. The goal is a shared language that lets an author build either a caravan camp or a starship maintenance shift with equal mechanical honesty and room for distinct fiction.

Back to C4AI v0.2 · Read the manifest